Heston Bot you’re a culinary genius! Neural networks have given birth to a new generation of chatbots that learn from their user. How are these next gen virtual assistants being used by fmcg?

Once consigned to the dustbin of technological history, along with Pac-Man and Betamax video, the virtual assistant is back. With the technology behind digital bots virtually unrecognisable since Microsoft debuted its Clippit office assistant in 1997, trilling ‘would you like help?’ no matter how many times you swore at it, a new swathe of brands and retailers are rolling out online assistants capable of doing everything from recommending fine wines, to providing recipes for coeliacs.

Five food and drink bots lending a hand

Heston Bot

Heston Bot
Catering to foodies desperate for a world class chef on hand in the kitchen, Heston Blumenthal teamed up with Microsoft in 2017 to launch the Heston Bot, available exclusively on Skype. Using the tech giant’s voice and messaging platforms, Heston Bot provides users with cooking tips, advice for esoteric flavour combinations based on his work, and also a few anecdotes. Advice contained in the bot is updated seasonally.

Lidl's margot

Lidl’s Margot
Aimed at helping shoppers overcome the ‘daunting process’ of selecting the right supermarket wine, Margot, which was developed by Aspect Software, is ‘a true wine enthusiast but definitely not snooty, loves food and company’. Available on Facebook Messenger, the bot provides information on how wine is made, tips on food pairing, or simply guides less confident shoppers to the right bottle for the right price.

Mars Perfect fit

Mars Perfect Fit Plus
Developed by ‘pet behaviourists, nutritionists and scientists’, the Mars bot, launched this month, helps pet owners create the ‘perfect’ program for their pets. Powered by chatbot platform Snaps, the bot can create a pet profile and then build a four-week scheme for ‘boosting your pet’s wellbeing’. At the end of this process users can also ask the bot specific questions about pet health.

M&S Christmas Conceirge

M&S Christmas Concierge
Launched in October 2017 and available until Christmas Eve, the Christmas Concierge was designed to help M&S shoppers better plan for the festive season by helping them to pick the right food products using the M&S range. Using Facebook Messenger to ask a series of questions, the bot helped consumers narrow choices for the festive menu, as well as offering tips on everything from hosting to presents.


Whole Foods Market recipe bot
Launched in 2016, in the wake of Facebook opening up its Messenger platform to businesses for the first time, the Whole Foods Market bot allows users to search for recipes, products and food inspiration, with options to filter by dietary restrictions or flavour preferences. Users can even search with an emoji, such as a jalapeño pepper or a banana, and receive a list of recipes containing that ingredient.

Crucially, this new generation of chatbots aims to make good on the failings of their predecessors. Where Clippit failed, say experts, was its inability to learn. The same action repeated over and over would be met with the same guileless idiocy. Its offers of help were often based on mistaken assumptions of what the user was trying to do and its remit was far too broad.

That’s all changed, explains Joe Smallman, technical lead at Cambridge Consultants. “You aren’t going to see a chatbot now that leaps in when you’re on a website and says ‘it looks like you’re trying to make a lasagne, would you like some help with that?’ These are programs that use neural networks to learn from examples so that they can become virtual assistants.”

A neural network is software that, when exposed to numerous behaviours, sees trends within them and changes its response accordingly. Designed to mimic the animal brain, it’s a far cry from previous attempts to create interaction between bots and humans.

“These chatbots try to work out what you’re saying and then construct a reply that is meaningful to you,” says Smallman. “Speech-based ones such as Siri, Google Assistant and Amazon’s Alexa all try to work out your intent.”

Move over Siri

It’s a skill that isn’t limited to tech giants though: some of the biggest brands in grocery are increasingly making use of. In October 2017, M&S launched its Christmas Concierge, which used the Facebook Messenger platform to deliver daily inspiration to shoppers on Christmas planning, with advice on menus and products.

“The ambition was to engage with people in a more personalised way and to raise awareness of the food range,” says Mathieu Abet, strategy director at Grey London, which developed the bot. That focus “plays to its strengths”, he adds “as the more single-minded the application is, the easier it is to interact with.” Users of the ‘concierge’, only available in the run up to Christmas, could simply swipe through on Facebook Messenger to see options for browsing menu ideas, recipes and the M&S range before clicking through to the website for further inspiration.

“It’s on Messenger, so you had to actively seek it out,” Abet continues. “This is important, because it isn’t intrusive and because it’s on a mobile app, there are different considerations. We wanted to keep it simple and limit how much you have to do. When you were asked questions, you tapped. It worked well and that makes us want to use our learning to understand what the next retail moment might be.”

Understanding retail moments and creating a tightly targeted chatbot that will exploit them seem to be the key to success. This way the chtbot need not master every food type, for example, but a narrower range relevant to its function. And therefore the time taken by each team of developers to equip the chatbot with relevant knowledge decreases. This was the case for the Christmas Concierge and it is also the thinking behind Margot, the chatbot launched by Lidl in January to sell its wine.

Like the M&S concierge, the Lidl bot knows its own precise retail moment - the notoriously tricky shopper mission of selecting the right wine. For shoppers it’s a “daunting process” says Alex Murray, digital director at Lidl UK, with the bot designed specifically to “help our customers become more confident in their wine knowledge”.


’They provide 24-hour service, answer the same questions and provide a basic level of help that is cheaper than a call centre’

“We called it Margot and gave it a personality with certain characteristics,” adds Tobias Goebel, director at Aspect Software, which developed the chatbot alongside Lidl. “We’re exploiting the effects of anthropomorphism as people love to associate human traits with something that seems human.” Like M&S, Lidl opted for Facebook Messenger as the platform for Margot as “it’s a conversational conduit already” adds Goebel.

WhatsApp opens up

The decision by the social media channel to open up its platform - and access to one billion users - to business in 2016 was one of the biggest drivers behind the new swathe of chatbots. Facebook-owned WhatsApp “will soon open up too” believes Goebel, while Apple announced a similar function for its iMessage in June 2017. All of which looks set to fuel the steady rise of the new generation of virtual assistants.

So advanced are some of the latest iterations that “even calling it a ‘bot’” undermines the complex functions it can fulfil, explains Derek Top, research director at San Francisco-based Opus Research. “Lidl is a perfect example, providing personalised, digital self-service 24 hours a day that makes your experience special, and makes you want to come back for more.”

And for businesses recruiting digital assistants over humans there are some clear advantages. “They provide 24-hour service, answer the same questions and provide a basic level of help that is going to be cheaper than running a call centre.”

Not that chatbots are without their challenges for shoppers. Of 3,000 UK and US consumers surveyed in late 2017 by Chatbots.org, set up in 2008 as a forum to share the latest developments in chatbot intelligence, 59% said they resented repeating information for chatbots, 53% rated them ‘not effective’ or ‘somewhat effective’, and 32% found that the chatbot had ‘got stuck and did not know what to do next’.

In fact, less than a year after granting business bots access to its Messenger platform, reports surfaced that 70% of chatbot launches on Facebook had failed, largely as only 30% of requests could be fulfilled without humans stepping in.


’People love to associate human traits with something that seems human’


The average age of users may play a role in their success, the Chatbots.org survey suggests, with young consumers reporting more fruitful interactions than older ones. Twenty-two per cent of Generation Z respondents rated their experience ‘very effective’.

Top isn’t deterred by negative feedback. “Whatever problems you have with customer service, this is technology having a conversation. There are so many different channels, from websites to phones, it creates so many more opportunities. This is just the start.”

For Smallman voice-activated mobile bots are the next frontier. “Research shows that most of the internet is viewed through your phone, which is a much more natural interface for conversation. And when you think about it, using a keyboard or a mouse to interact with a device seems strange. It will be obvious that you talk directly to your phone and that your phone acts as your virtual assistant.”

With each week that passes it seems another household name in grocery is inclined to agree as increasingly sophisticated chatbots are rolled out to lend a helping hand to customers, and huge sums of money and expertise are invested in their development.

In fact, so fast is the current pace of change it may well be that future generations look upon the likes of Margot and the M&S Christmas Concierge as a rudimentary technological cul-de-sac. Much like Clippit.